[Eeglablist] Is it possible to average independent components over groups?

[Marcel Prestel]

Dear list,

when working with independent components in clusters/groups, one will inevitably find that some subjects don’t seem to exhibit a certain component. Do any of you have experience with averaging components and applying them to other subjects? Or can anyone guide me to some information about it?

More concrete questions:
1) Does that make sense at all?
When averaging components we lose the main benefits of ICA such as independence and specificity for a single subject. Nevertheless, it could  be a good possibility to make comparisons over the whole group, especially when there are only few subjects missing the component.
2) How to normalize the components from the inverse weight matrix?
To make sure that each one contributes equally, the individual components have to be comparable. One could use z-scores, normalize the absolute values of the component vector to one, etc.

I have tried averaging components that have similar maps and show a distinct theta peak by normalizing the absolute values of the component vectors to one and taking the mean. When applied to both subjects with and without an individual component the theta peak was still discernible. This was a small group, but I think it indicates that the averaged component has „worked“.


Kind regards,
Marcel